Hui Li

Aaron Blakeman, Aarti Basant, Abhinav Khattar et al. · nvidia

As inference-time scaling becomes critical for enhanced reasoning capabilities, it is increasingly becoming important to build models that are efficient to infer. We introduce Nemotron-H, a family of 8B and 56B/47B hybrid Mamba-Transformer models designed to reduce inference cost for a given accuracy level. To achieve this goal, we replace the majority of self-attention layers in the common Transformer model architecture with Mamba layers that perform constant computation and require constant memory per generated token. We show that Nemotron-H models offer either better or on-par accuracy compared to other similarly-sized state-of-the-art open-sourced Transformer models (e.g., Qwen-2.5-7B/72B and Llama-3.1-8B/70B), while being up to 3$\times$ faster at inference. To further increase inference speed and reduce the memory required at inference time, we created Nemotron-H-47B-Base from the 56B model using a new compression via pruning and distillation technique called MiniPuzzle. Nemotron-H-47B-Base achieves similar accuracy to the 56B model, but is 20% faster to infer. In addition, we introduce an FP8-based training recipe and show that it can achieve on par results with BF16-based training. This recipe is used to train the 56B model. We are releasing Nemotron-H base model checkpoints with support in Hugging Face and NeMo.

Alberto Garcia-Duran, Roberto Gonzalez, Daniel Onoro-Rubio et al.

The text of a review expresses the sentiment a customer has towards a particular product. This is exploited in sentiment analysis where machine learning models are used to predict the review score from the text of the review. Furthermore, the products costumers have purchased in the past are indicative of the products they will purchase in the future. This is what recommender systems exploit by learning models from purchase information to predict the items a customer might be interested in. We propose TransRev, an approach to the product recommendation problem that integrates ideas from recommender systems, sentiment analysis, and multi-relational learning into a joint learning objective. TransRev learns vector representations for users, items, and reviews. The embedding of a review is learned such that (a) it performs well as input feature of a regression model for sentiment prediction; and (b) it always translates the reviewer embedding to the embedding of the reviewed items. This allows TransRev to approximate a review embedding at test time as the difference of the embedding of each item and the user embedding. The approximated review embedding is then used with the regression model to predict the review score for each item. TransRev outperforms state of the art recommender systems on a large number of benchmark data sets. Moreover, it is able to retrieve, for each user and item, the review text from the training set whose embedding is most similar to the approximated review embedding.